Computational thinking (CT) has emerged as a key topic of interest in K-12 education. Children that are exposed at an early age to STEM curriculum, such as computer programming and computational thinking, demonstrate fewer obstacles entering technical fields. Increased knowledge of programming and computation in early childhood is also associated with better problem solving, decision-making, basic number sense, language skills, and visual memory. As a digital competence, coding is explicitly regarded as a key 21st Century Skill, as the "literacy of today," such that its acquisition is regarded as essential to sustain economic development and competitiveness. Hence, the reliable evaluation of students' process data in context of problem solving tasks that require CT is of great importance. As opposed to product data, which only contain information about "what" the outcome of a problem solving process was (e.g. the final score), process data contain information about "how" the problem was solved (e.g. all the actions and problem solving steps). Students' coding processes are thus defined by their actions while coding, as evidenced by "process data," and are evaluated by comparing their action sequences to optimal action sequences. Prior research on process data analysis shows several inherent issues. Their approaches aggregate data and thus loses information which precludes them from being used in more detailed analyses of student behavior. Vector-based approaches often apply dimensionality reduction or normalization and require interpretation of the reduced dimensions, which is often not possible. Network-based or finite state visualizations that show transitions between states (i.e., actions or game-states), are aggregations over the student, game level, or time dimensions and thus lose detailed information along these dimensions. Additionally, these networks only model Markov processes of order one (current state and preceding state) and do not show the frequency of higher-order sequences such as transitions through more than one preceding state. Sequential pattern mining approaches can deal with higher-order sequences, but their results tend to be verbose and need tedious manual analysis. In summary, prior research has analyzed overall action sequences or code snapshots, but has not interpreted student actions in context of a situation during the problem solving process -- i.e. while students create the solution. A more fine-grained analysis of coding process data is needed, where relevant actions are interpreted as a part of the student's problem solving process. This paper addresses some of above issues and presents an approach to detect patterns related to computational thinking based on visualizations of students fine-grained actions in situational context.